Enable adam foreach in full train step tracing (#97897)
Main changes:
1. Registered several foreach ops to both meta and DTensor
2. Skip redundant getitem node when expanding foreach ops with DTensor
Pull Request resolved: https://github.com/pytorch/pytorch/pull/97897
Approved by: https://github.com/wanchaol, https://github.com/fegin
